RESEARCH OF TRAJECTORY TRACKING CONTROL FOR MOBILE ROBOT BASED ON REINFORCEMENT LEARNING TECHNIQUE

Abstract

Currently, the use of mobile robots is increasingly popular in industries. One of the important problems in motion control of mobile  robots is the control of tracking the reference motion trajectory. However, the mobile robot has a cascade control structure consisting of a dynamic controller in the inner ring and a kinematic controller in the outer ring. To solve the design problem without separating separate controllers, the paper presents a method using the online adaptive dynamic programming reinforcement learning technique with the structure using only a neural network approximating the function (OADP1NN). The algorithm can directly approximate the optimal solution (solution to the Hamilton Jacobi Bellman equation – HJB) simultaneously with the optimal control law. Performing simulations on Matlab software, the results showed that the OADP1NN algorithm has fully met two criteria for controlling  robots: tracking the reference trajectory and minimizing the cost function related to tracking error and control energy.